In a bid to strengthen India’s AI capabilities, the Government of India has selected Sarvam to build the country’s sovereign Large Language Model (LLM) under the ambitious IndiaAI Mission. The initiative marks the first time India will develop an indigenous foundational AI model from scratch, capable of reasoning, designed for voice, and fluent in Indian languages. The model is expected to be ready for secure, population-scale deployment, with the broader goal of promoting strategic autonomy.
Sarvam AI was chosen from among 67 shortlisted companies, marking a significant milestone in India's push for technological self-reliance.
“We are confident that Sarvam’s models will be competitive with global models,” said Shri Ashwini Vaishnaw, Union Minister for Electronics & Information Technology, Railways, and Information & Broadcasting.
Sarvam AI is at the forefront of creating AI models tailored to India's linguistic and cultural diversity. The company's flagship model, Sarvam-1, is a 2-billion-parameter LLM trained entirely within India, utilizing a dataset of 4 trillion tokens. This model supports ten major Indian languages—Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada, and Bengali—and has outperformed global counterparts like Google's Gemma-2 and Meta's Llama-3 on certain benchmarks.
Sarvam AI's approach emphasizes sovereign AI development, ensuring that data remains within national borders and adheres to local laws and regulations. The company has also launched Sarvam Agents, voice-enabled, multilingual business agents available via telephone, WhatsApp, or in-app interfaces, and Shuka 1.0, India's first open-source AudioLM supporting Indian language voice input.
Also Read: Sarvam AI is building India’s first homegrown Generative AI for every language
